Thomas M Kehrenberg

We already have a dataset with non-binary target (NonBinaryToy) and several with non-binary groups (for example adult with race), but AFAIK no dataset with both.

This is a prerequisite for using the tmpdir fixture in the tests

The great renaming

4

## Remove the lowercase aliases for the dataset classes - [x] `adult()` -> *deleted* - [x] `compas()` -> *deleted* ## Algorithm renames - [x] `Kamiran` -> `Reweighting` – done in...

Avoid division by zero in TPR (and similar metrics)

3

TPR is defined like this: ```python return t_pos / (t_pos + f_neg) ``` If both `t_pos` and `f_neg` are 0, then numpy produces `NaN`. But I'd argue that the result...

Polars seems faster and it seems to have been designed with static type checking in mind, which you definitely cannot say about Pandas.

this needs tidying up a bit. Need to think a bit about the best way to deal with discrete predictions. (Or if they need to be handled at all) 

the idea is that we have one LAFTR model in the implementations dir, then two LAFTR models in the algorithms dir, one in the inprocess (this one) which is supervised,...

See how algofairness does it: ```python class DIBinary(Metric): def __init__(self): # ... def calc(self, actual, predicted, dict_of_sensitive_lists, single_sensitive_name, unprotected_vals, positive_pred): # ... return DI def is_better_than(self, val1, val2): dist1 =...

https://meps.ahrq.gov/mepsweb/